add neon opts for matrix procs

git-svn-id: http://skia.googlecode.com/svn/trunk@353 2bbb7eff-a529-9590-31e7-b0007b416f81

6 files changed, 1647 insertions, 8 deletions

diff --git a/include/core/SkPaint.h b/include/core/SkPaint.h
index ed60a20f61..659d8753e7 100644
--- a/include/core/SkPaint.h
+++ b/include/core/SkPaint.h
@@ -256,11 +256,16 @@ public:
     /** Styles apply to rect, oval, path, and text.
         Bitmaps are always drawn in "fill", and lines are always drawn in
         "stroke".
+     
+        Note: strokeandfill implicitly draws the result with
+        SkPath::kWinding_FillType, so if the original path is even-odd, the
+        results may not appear the same as if it was drawn twice, filled and
+        then stroked.
     */
     enum Style {
-        kFill_Style,            //!< fill with the paint's color
-        kStroke_Style,          //!< stroke with the paint's color
-        kStrokeAndFill_Style,   //!< fill and stroke with the paint's color
+        kFill_Style,            //!< fill the geometry
+        kStroke_Style,          //!< stroke the geometry
+        kStrokeAndFill_Style,   //!< fill and stroke the geometry
 
         kStyleCount,
     };
diff --git a/samplecode/SampleStrokePath.cpp b/samplecode/SampleStrokePath.cpp
new file mode 100644
index 0000000000..ec0a970e94
--- /dev/null
+++ b/samplecode/SampleStrokePath.cpp
@@ -0,0 +1,125 @@
+#include "SampleCode.h"
+#include "SkCanvas.h"
+#include "SkParsePath.h"
+#include "SkPath.h"
+#include "SkRandom.h"
+#include "SkView.h"
+
+static void scale_to_width(SkPath* path, SkScalar dstWidth) {
+    const SkRect& bounds = path->getBounds();
+    SkScalar scale = dstWidth / bounds.width();
+    SkMatrix matrix;
+
+    matrix.setScale(scale, scale);
+    path->transform(matrix);
+}
+
+static const struct {
+    SkPaint::Style  fStyle;
+    SkPaint::Join   fJoin;
+    int             fStrokeWidth;
+} gRec[] = {
+    { SkPaint::kFill_Style,             SkPaint::kMiter_Join,   0 },
+    { SkPaint::kStroke_Style,           SkPaint::kMiter_Join,   0 },
+    { SkPaint::kStroke_Style,           SkPaint::kMiter_Join,   10 },
+    { SkPaint::kStrokeAndFill_Style,    SkPaint::kMiter_Join,   10 },
+};
+
+class StrokePathView : public SkView {
+    SkScalar    fWidth;
+    SkPath      fPath;
+public:
+	StrokePathView() {
+        fWidth = SkIntToScalar(120);
+
+#if 0
+        const char str[] =
+            "M 0, 3"
+            "C 10, -10, 30, -10, 0, 28"
+            "C -30, -10, -10, -10, 0, 3"
+            "Z";
+        SkParsePath::FromSVGString(str, &fPath);
+#else
+        fPath.addCircle(0, 0, SkIntToScalar(50), SkPath::kCW_Direction);
+        fPath.addCircle(0, SkIntToScalar(-50), SkIntToScalar(30), SkPath::kCW_Direction);
+#endif
+        
+        scale_to_width(&fPath, fWidth);
+        const SkRect& bounds = fPath.getBounds();
+        fPath.offset(-bounds.fLeft, -bounds.fTop);
+    }
+    
+protected:
+    // overrides from SkEventSink
+    virtual bool onQuery(SkEvent* evt) {
+        if (SampleCode::TitleQ(*evt)) {
+            SampleCode::TitleR(evt, "StrokePath");
+            return true;
+        }
+        return this->INHERITED::onQuery(evt);
+    }
+    
+    void drawBG(SkCanvas* canvas) {
+        canvas->drawColor(0xFFDDDDDD);
+    }
+    
+    SkRandom rand;
+    
+    void drawSet(SkCanvas* canvas, SkPaint* paint) {
+        SkAutoCanvasRestore acr(canvas, true);
+
+        for (size_t i = 0; i < SK_ARRAY_COUNT(gRec); i++) {
+            paint->setStyle(gRec[i].fStyle);
+            paint->setStrokeJoin(gRec[i].fJoin);
+            paint->setStrokeWidth(SkIntToScalar(gRec[i].fStrokeWidth));
+            canvas->drawPath(fPath, *paint);
+            canvas->translate(fWidth * 5 / 4, 0);
+        }
+    }
+
+    virtual void onDraw(SkCanvas* canvas) {
+        drawBG(canvas);
+        canvas->translate(SkIntToScalar(10), SkIntToScalar(10));
+
+        SkPaint paint;
+        paint.setAntiAlias(true);
+        paint.setColor(SK_ColorBLUE);
+
+#if 1
+        SkPath p;
+        float r = rand.nextUScalar1() + 0.5f;
+        SkScalar x = 0, y = 0;
+        p.moveTo(x, y);
+#if 0
+        p.cubicTo(x-75*r, y+75*r, x-40*r, y+125*r, x, y+85*r);
+        p.cubicTo(x+40*r, y+125*r, x+75*r, y+75*r, x, y);
+#else
+        p.cubicTo(x+75*r, y+75*r, x+40*r, y+125*r, x, y+85*r);
+        p.cubicTo(x-40*r, y+125*r, x-75*r, y+75*r, x, y);
+#endif
+        p.close();
+        fPath = p;
+        fPath.offset(100, 0);
+#endif
+        
+        fPath.setFillType(SkPath::kWinding_FillType);
+        drawSet(canvas, &paint);
+        
+        canvas->translate(0, fPath.getBounds().height() * 5 / 4);
+        fPath.setFillType(SkPath::kEvenOdd_FillType);
+        drawSet(canvas, &paint);
+    }
+
+    virtual SkView::Click* onFindClickHandler(SkScalar x, SkScalar y) {
+        this->inval(NULL);
+        return this->INHERITED::onFindClickHandler(x, y);
+    }
+private:
+    typedef SkView INHERITED;
+};
+
+//////////////////////////////////////////////////////////////////////////////
+
+static SkView* MyFactory() { return new StrokePathView; }
+static SkViewRegister reg(MyFactory);
+
diff --git a/src/core/SkBitmapProcState_matrixProcs.cpp b/src/core/SkBitmapProcState_matrixProcs.cpp
index 67b57c4418..6654312a8f 100644
--- a/src/core/SkBitmapProcState_matrixProcs.cpp
+++ b/src/core/SkBitmapProcState_matrixProcs.cpp
@@ -1,3 +1,5 @@
+/* NEON optimized code (C) COPYRIGHT 2009 Motorola */
+
 #include "SkBitmapProcState.h"
 #include "SkPerspIter.h"
 #include "SkShader.h"
@@ -31,14 +33,22 @@ void decal_filter_scale(uint32_t dst[], SkFixed fx, SkFixed dx, int count);
 #define TILEX_LOW_BITS(fx, max) (((fx) >> 12) & 0xF)
 #define TILEY_LOW_BITS(fy, max) (((fy) >> 12) & 0xF)
 #define CHECK_FOR_DECAL
-#include "SkBitmapProcState_matrix.h"
+#if	defined(__ARM_HAVE_NEON)
+    #include "SkBitmapProcState_matrix_clamp.h"
+#else
+    #include "SkBitmapProcState_matrix.h"
+#endif
 
 #define MAKENAME(suffix)        RepeatX_RepeatY ## suffix
 #define TILEX_PROCF(fx, max)    (((fx) & 0xFFFF) * ((max) + 1) >> 16)
 #define TILEY_PROCF(fy, max)    (((fy) & 0xFFFF) * ((max) + 1) >> 16)
 #define TILEX_LOW_BITS(fx, max) ((((fx) & 0xFFFF) * ((max) + 1) >> 12) & 0xF)
 #define TILEY_LOW_BITS(fy, max) ((((fy) & 0xFFFF) * ((max) + 1) >> 12) & 0xF)
-#include "SkBitmapProcState_matrix.h"
+#if	defined(__ARM_HAVE_NEON)
+    #include "SkBitmapProcState_matrix_repeat.h"
+#else
+    #include "SkBitmapProcState_matrix.h"
+#endif
 
 #define MAKENAME(suffix)        GeneralXY ## suffix
 #define PREAMBLE(state)         SkBitmapProcState::FixedTileProc tileProcX = (state).fTileProcX; \
@@ -147,6 +157,52 @@ void decal_nofilter_scale(uint32_t dst[], SkFixed fx, SkFixed dx, int count)
 {
     int i;
 
+#if	defined(__ARM_HAVE_NEON)
+    if (count >= 8) {
+        /* SkFixed is 16.16 fixed point */
+        SkFixed dx2 = dx+dx;
+        SkFixed dx4 = dx2+dx2;
+        SkFixed dx8 = dx4+dx4;
+
+        /* now build fx/fx+dx/fx+2dx/fx+3dx */
+        SkFixed fx1, fx2, fx3;
+        int32x2_t lower, upper;
+        int32x4_t lbase, hbase;
+        uint16_t *dst16 = (uint16_t *)dst;
+
+        fx1 = fx+dx;
+        fx2 = fx1+dx;
+        fx3 = fx2+dx;
+
+        /* avoid an 'lbase unitialized' warning */
+        lbase = vdupq_n_s32(fx);
+        lbase = vsetq_lane_s32(fx1, lbase, 1);
+        lbase = vsetq_lane_s32(fx2, lbase, 2);
+        lbase = vsetq_lane_s32(fx3, lbase, 3);
+        hbase = vaddq_s32(lbase, vdupq_n_s32(dx4));
+
+        /* take upper 16 of each, store, and bump everything */
+        do {
+            int32x4_t lout, hout;
+            uint16x8_t hi16;
+
+            lout = lbase;
+            hout = hbase;
+            /* gets hi's of all louts then hi's of all houts */
+            asm ("vuzpq.16 %q0, %q1" : "+w" (lout), "+w" (hout));
+            hi16 = vreinterpretq_u16_s32(hout);
+            vst1q_u16(dst16, hi16);
+
+            /* on to the next */
+            lbase = vaddq_s32 (lbase, vdupq_n_s32(dx8));
+            hbase = vaddq_s32 (hbase, vdupq_n_s32(dx8));
+            dst16 += 8;
+            count -= 8;
+            fx += dx8;
+        } while (count >= 8);
+        dst = (uint32_t *) dst16;
+    }
+#else
     for (i = (count >> 2); i > 0; --i)
     {
         *dst++ = pack_two_shorts(fx >> 16, (fx + dx) >> 16);
@@ -154,16 +210,55 @@ void decal_nofilter_scale(uint32_t dst[], SkFixed fx, SkFixed dx, int count)
         *dst++ = pack_two_shorts(fx >> 16, (fx + dx) >> 16);
         fx += dx+dx;
     }
-    uint16_t* xx = (uint16_t*)dst;
+    count &= 3;
+#endif
 
-    for (i = (count & 3); i > 0; --i)
-    {
+    uint16_t* xx = (uint16_t*)dst;
+    for (i = count; i > 0; --i) {
         *xx++ = SkToU16(fx >> 16); fx += dx;
     }
 }
 
 void decal_filter_scale(uint32_t dst[], SkFixed fx, SkFixed dx, int count)
 {
+
+#if	defined(__ARM_HAVE_NEON)
+    if (count >= 8) {
+        int32x4_t wide_fx;
+        int32x4_t wide_fx2;
+        int32x4_t wide_dx8 = vdupq_n_s32(dx*8);
+
+        wide_fx = vdupq_n_s32(fx);
+        wide_fx = vsetq_lane_s32(fx+dx, wide_fx, 1);
+        wide_fx = vsetq_lane_s32(fx+dx+dx, wide_fx, 2);
+        wide_fx = vsetq_lane_s32(fx+dx+dx+dx, wide_fx, 3);
+
+        wide_fx2 = vaddq_s32(wide_fx, vdupq_n_s32(dx+dx+dx+dx));
+
+        while (count >= 8) {
+            int32x4_t wide_out;
+            int32x4_t wide_out2;
+
+            wide_out = vshlq_n_s32(vshrq_n_s32(wide_fx, 12), 14);
+            wide_out = vorrq_s32(wide_out,
+            vaddq_s32(vshrq_n_s32(wide_fx,16), vdupq_n_s32(1)));
+
+            wide_out2 = vshlq_n_s32(vshrq_n_s32(wide_fx2, 12), 14);
+            wide_out2 = vorrq_s32(wide_out2,
+            vaddq_s32(vshrq_n_s32(wide_fx2,16), vdupq_n_s32(1)));
+
+            vst1q_u32(dst, vreinterpretq_u32_s32(wide_out));
+            vst1q_u32(dst+4, vreinterpretq_u32_s32(wide_out2));
+
+            dst += 8;
+            fx += dx*8;
+            wide_fx = vaddq_s32(wide_fx, wide_dx8);
+            wide_fx2 = vaddq_s32(wide_fx2, wide_dx8);
+            count -= 8;
+        }
+    }
+#endif
+
     if (count & 1)
     {
         SkASSERT((fx >> (16 + 14)) == 0);
diff --git a/src/core/SkBitmapProcState_matrix_clamp.h b/src/core/SkBitmapProcState_matrix_clamp.h
new file mode 100644
index 0000000000..c95a4fa065
--- /dev/null
+++ b/src/core/SkBitmapProcState_matrix_clamp.h
@@ -0,0 +1,826 @@
+/* NEON optimized code (C) COPYRIGHT 2009 Motorola */
+/*
+ * Modifications done in-house at Motorola 
+ *
+ * this is a clone of SkBitmapProcState_matrix.h
+ * and has been tuned to work with the NEON unit.
+ *
+ * Still going back and forth between whether this approach
+ * (clone the entire SkBitmapProcState_matrix.h file or
+ * if I should put just the modified routines in here and
+ * then use a construct like #define DONT_DO_THIS_FUNCTION or
+ * something like that...
+ *
+ * This is for the ClampX_ClampY instance
+ *
+ */
+
+
+#if	!defined(__ARM_HAVE_NEON)
+#error	this file can be used only when the NEON unit is enabled
+#endif
+
+#include <arm_neon.h>
+
+/*
+ * This has been modified on the knowledge that (at the time)
+ * we had the following macro definitions in the parent file
+ *
+ * #define MAKENAME(suffix)        ClampX_ClampY ## suffix
+ * #define TILEX_PROCF(fx, max)    SkClampMax((fx) >> 16, max)
+ * #define TILEY_PROCF(fy, max)    SkClampMax((fy) >> 16, max)
+ * #define TILEX_LOW_BITS(fx, max) (((fx) >> 12) & 0xF)
+ * #define TILEY_LOW_BITS(fy, max) (((fy) >> 12) & 0xF)
+ * #define CHECK_FOR_DECAL
+ */
+
+/* SkClampMax(val,max) -- bound to 0..max */
+
+#define SCALE_NOFILTER_NAME     MAKENAME(_nofilter_scale)
+#define SCALE_FILTER_NAME       MAKENAME(_filter_scale)
+#define AFFINE_NOFILTER_NAME    MAKENAME(_nofilter_affine)
+#define AFFINE_FILTER_NAME      MAKENAME(_filter_affine)
+#define PERSP_NOFILTER_NAME     MAKENAME(_nofilter_persp)
+#define PERSP_FILTER_NAME       MAKENAME(_filter_persp)
+
+#define PACK_FILTER_X_NAME  MAKENAME(_pack_filter_x)
+#define PACK_FILTER_Y_NAME  MAKENAME(_pack_filter_y)
+
+#ifndef PREAMBLE
+    #define PREAMBLE(state)
+    #define PREAMBLE_PARAM_X
+    #define PREAMBLE_PARAM_Y
+    #define PREAMBLE_ARG_X
+    #define PREAMBLE_ARG_Y
+#endif
+
+static void SCALE_NOFILTER_NAME(const SkBitmapProcState& s,
+                                uint32_t xy[], int count, int x, int y) {
+    SkASSERT((s.fInvType & ~(SkMatrix::kTranslate_Mask |
+                             SkMatrix::kScale_Mask)) == 0);
+
+    PREAMBLE(s);
+    // we store y, x, x, x, x, x
+
+    const unsigned maxX = s.fBitmap->width() - 1;
+    SkFixed fx;
+    {
+        SkPoint pt;
+        s.fInvProc(*s.fInvMatrix, SkIntToScalar(x) + SK_ScalarHalf,
+                                  SkIntToScalar(y) + SK_ScalarHalf, &pt);
+        fx = SkScalarToFixed(pt.fY);
+        const unsigned maxY = s.fBitmap->height() - 1;
+        *xy++ = TILEY_PROCF(fx, maxY);
+        fx = SkScalarToFixed(pt.fX);
+    }
+    
+    if (0 == maxX) {
+        // all of the following X values must be 0
+        memset(xy, 0, count * sizeof(uint16_t));
+        return;
+    }
+
+    const SkFixed dx = s.fInvSx;
+
+#ifdef CHECK_FOR_DECAL
+    // test if we don't need to apply the tile proc
+    if ((unsigned)(fx >> 16) <= maxX &&
+        (unsigned)((fx + dx * (count - 1)) >> 16) <= maxX) {
+        decal_nofilter_scale(xy, fx, dx, count);
+        return;
+    }
+#endif
+
+    int i;
+
+	/* very much like done in decal_nofilter, but with
+	 * an extra clamping function applied.
+	 * TILEX_PROCF(fx,max) SkClampMax((fx)>>16, max)
+	 */
+	if (count >= 8) {
+	    /* SkFixed is 16.16 fixed point */
+	    SkFixed dx2 = dx+dx;
+	    SkFixed dx4 = dx2+dx2;
+	    SkFixed dx8 = dx4+dx4;
+
+	    /* now build fx/fx+dx/fx+2dx/fx+3dx */
+	    SkFixed fx1, fx2, fx3;
+	    int32x2_t lower, upper;
+	    int32x4_t lbase, hbase;
+	    int16_t *dst16 = (int16_t *)xy;
+
+	    fx1 = fx+dx;
+	    fx2 = fx1+dx;
+	    fx3 = fx2+dx;
+
+	    /* build my template(s) */
+	    /* avoid the 'lbase unitialized' warning */
+	    lbase = vdupq_n_s32(fx);
+	    lbase = vsetq_lane_s32(fx1, lbase, 1);
+	    lbase = vsetq_lane_s32(fx2, lbase, 2);
+	    lbase = vsetq_lane_s32(fx3, lbase, 3);
+
+	    hbase = vaddq_s32(lbase, vdupq_n_s32(dx4));
+
+	    /* store & bump */
+	    do {
+	        int32x4_t lout;
+            int32x4_t hout;
+            int16x8_t hi16;
+
+            /* get the hi 16s of all those 32s */
+            lout = lbase;
+            hout = hbase;
+            /* this sets up all lout's then all hout's in hout */
+            asm ("vuzpq.16 %q0, %q1" : "+w" (lout), "+w" (hout));
+            hi16 = vreinterpretq_s16_s32(hout);
+
+            /* clamp & output */
+            hi16 = vmaxq_s16(hi16, vdupq_n_s16(0));
+            hi16 = vminq_s16(hi16, vdupq_n_s16(maxX));
+            vst1q_s16(dst16, hi16);
+
+            /* but preserving base & on to the next */
+            lbase = vaddq_s32 (lbase, vdupq_n_s32(dx8));
+            hbase = vaddq_s32 (hbase, vdupq_n_s32(dx8));
+            dst16 += 8;
+            count -= 8;
+            fx += dx8;
+	    } while (count >= 8);
+	    xy = (uint32_t *) dst16;
+	}
+
+    uint16_t* xx = (uint16_t*)xy;
+    for (i = count; i > 0; --i) {
+        *xx++ = TILEX_PROCF(fx, maxX); fx += dx;
+    }
+}
+
+// note: we could special-case on a matrix which is skewed in X but not Y.
+// this would require a more general setup thatn SCALE does, but could use
+// SCALE's inner loop that only looks at dx
+
+static void AFFINE_NOFILTER_NAME(const SkBitmapProcState& s,
+                                 uint32_t xy[], int count, int x, int y) {
+    SkASSERT(s.fInvType & SkMatrix::kAffine_Mask);
+    SkASSERT((s.fInvType & ~(SkMatrix::kTranslate_Mask |
+                             SkMatrix::kScale_Mask |
+                             SkMatrix::kAffine_Mask)) == 0);
+    
+    PREAMBLE(s);
+    SkPoint srcPt;
+    s.fInvProc(*s.fInvMatrix,
+               SkIntToScalar(x) + SK_ScalarHalf,
+               SkIntToScalar(y) + SK_ScalarHalf, &srcPt);
+    
+    SkFixed fx = SkScalarToFixed(srcPt.fX);
+    SkFixed fy = SkScalarToFixed(srcPt.fY);
+    SkFixed dx = s.fInvSx;
+    SkFixed dy = s.fInvKy;
+    int maxX = s.fBitmap->width() - 1;
+    int maxY = s.fBitmap->height() - 1;
+    
+    /* NEON lets us do an 8x unrolling */
+    if (count >= 8) {
+        /* SkFixed is 16.16 fixed point */
+        SkFixed dx4 = dx * 4;
+        SkFixed dy4 = dy * 4;
+        SkFixed dx8 = dx * 8;
+        SkFixed dy8 = dy * 8;
+
+        int32x4_t xbase, ybase;
+        int32x4_t x2base, y2base;
+        int16_t *dst16 = (int16_t *) xy;
+
+        /* my sets of maxx/maxy for clamping */
+        int32_t maxpair = (maxX&0xffff) | ((maxY&0xffff)<<16);
+        int16x8_t maxXY = vreinterpretq_s16_s32(vdupq_n_s32(maxpair));
+
+        /* now build fx/fx+dx/fx+2dx/fx+3dx */
+        /* avoid the 'xbase unitialized' warning...*/
+        xbase = vdupq_n_s32(fx);
+        xbase = vsetq_lane_s32(fx+dx, xbase, 1);
+        xbase = vsetq_lane_s32(fx+dx+dx, xbase, 2);
+        xbase = vsetq_lane_s32(fx+dx+dx+dx, xbase, 3);
+
+        /* same for fy */
+        /* avoid the 'ybase unitialized' warning...*/
+        ybase = vdupq_n_s32(fy);
+        ybase = vsetq_lane_s32(fy+dy, ybase, 1);
+        ybase = vsetq_lane_s32(fy+dy+dy, ybase, 2);
+        ybase = vsetq_lane_s32(fy+dy+dy+dy, ybase, 3);
+
+        x2base = vaddq_s32(xbase, vdupq_n_s32(dx4));
+        y2base = vaddq_s32(ybase, vdupq_n_s32(dy4));
+
+        /* store & bump */
+        do {
+            int32x4_t xout, yout;
+            int32x4_t x2out, y2out;
+            int16x8_t hi16, hi16_2;
+
+            xout = xbase;
+            yout = ybase;
+
+            /* overlay y's low16 with hi16 from x */
+            /* so we properly shifted xyxyxyxy */
+            yout = vsriq_n_s32(yout, xout, 16);
+            hi16 = vreinterpretq_s16_s32 (yout);
+
+            /* do the clamping; both guys get 0's */
+            hi16 = vmaxq_s16 (hi16, vdupq_n_s16(0));
+            hi16 = vminq_s16 (hi16, maxXY);
+
+            vst1q_s16 (dst16, hi16);
+
+            /* and for the other 4 pieces of this iteration */
+            x2out = x2base;
+            y2out = y2base;
+
+            /* overlay y's low16 with hi16 from x */
+            /* so we properly shifted xyxyxyxy */
+            y2out = vsriq_n_s32(y2out, x2out, 16);
+            hi16_2 = vreinterpretq_s16_s32 (y2out);
+
+            /* do the clamping; both guys get 0's */
+            hi16_2 = vmaxq_s16 (hi16_2, vdupq_n_s16(0));
+            hi16_2 = vminq_s16 (hi16_2, maxXY);
+
+            /* RBE: gcc regenerates dst16+8 all the time instead
+             * of folding it into an addressing mode. *sigh* */
+            vst1q_s16 (dst16+8, hi16_2);
+
+            /* moving base and on to the next */
+            xbase = vaddq_s32 (xbase, vdupq_n_s32 (dx8));
+            ybase = vaddq_s32 (ybase, vdupq_n_s32 (dy8));
+            x2base = vaddq_s32 (x2base, vdupq_n_s32 (dx8));
+            y2base = vaddq_s32 (y2base, vdupq_n_s32 (dy8));
+
+            dst16 += 16;		/* 8x32 aka 16x16 */
+            count -= 8;
+            fx += dx8;
+            fy += dy8;
+        } while (count >= 8);
+        xy = (uint32_t *) dst16;
+    }
+
+    for (int i = count; i > 0; --i) {
+        *xy++ = (TILEY_PROCF(fy, maxY) << 16) | TILEX_PROCF(fx, maxX);
+        fx += dx; fy += dy;
+    }
+}
+
+static void PERSP_NOFILTER_NAME(const SkBitmapProcState& s,
+                                uint32_t* SK_RESTRICT xy,
+                                int count, int x, int y) {
+    SkASSERT(s.fInvType & SkMatrix::kPerspective_Mask);
+    
+    PREAMBLE(s);
+    /* max{X,Y} are int here, but later shown/assumed to fit in 16 bits */
+    int maxX = s.fBitmap->width() - 1;
+    int maxY = s.fBitmap->height() - 1;
+    
+    SkPerspIter   iter(*s.fInvMatrix,
+                       SkIntToScalar(x) + SK_ScalarHalf,
+                       SkIntToScalar(y) + SK_ScalarHalf, count);
+    
+    while ((count = iter.next()) != 0) {
+        const SkFixed* SK_RESTRICT srcXY = iter.getXY();
+
+        /* srcXY is a batch of 32 bit numbers X0,Y0,X1,Y1...
+         * but we immediately discard the low 16 bits...
+         * so what we're going to do is vld4, which will give us
+         * xlo,xhi,ylo,yhi distribution and we can ignore the 'lo'
+         * parts....
+         */
+        if (count >= 8) {
+            int16_t *mysrc = (int16_t *) srcXY;
+            int16_t *mydst = (int16_t *) xy;
+            int16x4_t maxX4 = vdup_n_s16((int16_t)maxX);
+            int16x4_t maxY4 = vdup_n_s16((int16_t)maxY);
+            int16x4_t zero4 = vdup_n_s16(0);
+            do {
+                register int16x4_t xlo asm("d0");
+                register int16x4_t xhi asm("d1");
+                register int16x4_t ylo asm("d2");
+                register int16x4_t yhi asm("d3");
+                register int16x4_t x2lo asm("d4");
+                register int16x4_t x2hi asm("d5");
+                register int16x4_t y2lo asm("d6");
+                register int16x4_t y2hi asm("d7");
+
+                register int16x4_t out_xhi asm("d8");
+                register int16x4_t out_yhi asm("d9");
+                register int16x4_t out_x2hi asm("d10");
+                register int16x4_t out_y2hi asm("d11");
+
+
+                /* vld4 does the de-interleaving for us */
+                /* dependent on register assignments above */
+                    asm ("vld4.16	{d0-d3},[%4]  /* xlo=%P0 xhi=%P1 ylo=%P2 yhi=%P3 */"
+                : "=w" (xlo), "=w" (xhi), "=w" (ylo), "=w" (yhi)
+                : "r" (mysrc)
+                );
+
+                /* offset == 256 bits == 32 bytes == 8 longs */
+                    asm ("vld4.16	{d4-d7},[%4,#32]  /* xlo=%P0 xhi=%P1 ylo=%P2 yhi=%P3 */"
+                : "=w" (x2lo), "=w" (x2hi), "=w" (y2lo), "=w" (y2hi)
+                : "r" (mysrc)
+                );
+
+                /* clamp the first 4 here */
+
+                /* clamp X>>16 (aka xhi) to 0..maxX */
+                xhi = vmax_s16(xhi, zero4);	/* now 0.. */
+                out_xhi = vmin_s16(xhi, maxX4);	/* now 0..maxX */
+
+                /* clamp Y>>16 (aka yhi) to 0..maxY */
+                yhi = vmax_s16(yhi, zero4);	/* now 0.. */
+                out_yhi = vmin_s16(yhi, maxY4);	/* now 0..maxY */
+
+                /* clamp the second 4 here */
+
+                /* clamp X>>16 (aka xhi) to 0..maxX */
+                x2hi = vmax_s16(x2hi, zero4);	/* now 0.. */
+                out_x2hi = vmin_s16(x2hi, maxX4);	/* now 0..maxX */
+
+                /* clamp Y>>16 (aka yhi) to 0..maxY */
+                y2hi = vmax_s16(y2hi, zero4);	/* now 0.. */
+                out_y2hi = vmin_s16(y2hi, maxY4);	/* now 0..maxY */
+
+                /* we're storing as {x,y}s: x is [0], y is [1] */
+                /* we'll use vst2 to make this happen */
+                /* XXX: could use auto increment! */
+                    asm ("vst2.16	{d8-d9},[%2]  /* xlo=%P0 xhi=%P1 */"
+                :
+                : "w" (out_xhi), "w" (out_yhi), "r" (mydst)
+                );
+
+                /* offset == 16 bytes == 8 shorts */
+                    asm ("vst2.16	{d10-d11},[%2,#16]  /* xlo=%P0 xhi=%P1 */"
+                :
+                : "w" (out_x2hi), "w" (out_y2hi), "r" (mydst)
+                );
+
+                /* XXX: gcc isn't interleaving these with the NEON ops
+                 * but i think that all the scoreboarding works out */
+                count -= 8;	/* 8 iterations */
+                mysrc += 32;	/* 16 longs, aka 32 shorts */
+                mydst += 16;	/* 16 shorts, aka 8 longs */
+            } while (count >= 8);
+            /* get xy and srcXY fixed up */
+            srcXY = (const SkFixed *) mysrc;
+            xy = (uint32_t *) mydst;
+        }
+
+        while (--count >= 0) {
+            *xy++ = (TILEY_PROCF(srcXY[1], maxY) << 16) |
+                     TILEX_PROCF(srcXY[0], maxX);
+            srcXY += 2;
+        }
+    }
+}
+
+//////////////////////////////////////////////////////////////////////////////
+
+static inline uint32_t PACK_FILTER_Y_NAME(SkFixed f, unsigned max,
+                                          SkFixed one PREAMBLE_PARAM_Y) {
+    unsigned i = TILEY_PROCF(f, max);
+    i = (i << 4) | TILEY_LOW_BITS(f, max);
+    return (i << 14) | (TILEY_PROCF((f + one), max));
+}
+
+static inline uint32_t PACK_FILTER_X_NAME(SkFixed f, unsigned max,
+                                          SkFixed one PREAMBLE_PARAM_X) {
+    unsigned i = TILEX_PROCF(f, max);
+    i = (i << 4) | TILEX_LOW_BITS(f, max);
+    return (i << 14) | (TILEX_PROCF((f + one), max));
+}
+
+static void SCALE_FILTER_NAME(const SkBitmapProcState& s,
+                              uint32_t xy[], int count, int x, int y) {
+    SkASSERT((s.fInvType & ~(SkMatrix::kTranslate_Mask |
+                             SkMatrix::kScale_Mask)) == 0);
+    SkASSERT(s.fInvKy == 0);
+
+    PREAMBLE(s);
+    
+    const unsigned maxX = s.fBitmap->width() - 1;
+    const SkFixed one = s.fFilterOneX;
+    const SkFixed dx = s.fInvSx;
+    SkFixed fx;
+
+    {
+        SkPoint pt;
+        s.fInvProc(*s.fInvMatrix, SkIntToScalar(x) + SK_ScalarHalf,
+                                  SkIntToScalar(y) + SK_ScalarHalf, &pt);
+        const SkFixed fy = SkScalarToFixed(pt.fY) - (s.fFilterOneY >> 1);
+        const unsigned maxY = s.fBitmap->height() - 1;
+        // compute our two Y values up front
+        *xy++ = PACK_FILTER_Y_NAME(fy, maxY, s.fFilterOneY PREAMBLE_ARG_Y);
+        // now initialize fx
+        fx = SkScalarToFixed(pt.fX) - (one >> 1);
+    }
+
+#ifdef CHECK_FOR_DECAL
+    // test if we don't need to apply the tile proc
+    if (dx > 0 &&
+            (unsigned)(fx >> 16) <= maxX &&
+            (unsigned)((fx + dx * (count - 1)) >> 16) < maxX) {
+        decal_filter_scale(xy, fx, dx, count);
+    } else
+#endif
+
+    if (count >= 4) {
+        int32x4_t wide_dx, wide_one;
+        int32x4_t wide_fx, wide_fx1, wide_i, wide_lo;
+    #if 0
+        /* verification hooks -- see below */
+        SkFixed debug_fx = fx;
+        int count_done = 0;
+    #endif
+
+        wide_fx = vdupq_n_s32(fx);
+        wide_fx = vsetq_lane_s32(fx+dx, wide_fx, 1);
+        wide_fx = vsetq_lane_s32(fx+dx+dx, wide_fx, 2);
+        wide_fx = vsetq_lane_s32(fx+dx+dx+dx, wide_fx, 3);
+
+        wide_dx = vdupq_n_s32(dx);
+        wide_one = vdupq_n_s32(one);
+
+        while (count >= 4) {
+            /* original expands to: 
+             * unsigned i = SkClampMax((f) >> 16, max);
+             * i = (i << 4) | (((f) >> 12) & 0xF);
+             * return (i << 14) | (SkClampMax(((f + one)) >> 16, max));
+             */
+
+            /* i = SkClampMax(f>>16, maxX) */
+            wide_i = vmaxq_s32(vshrq_n_s32(wide_fx,16), vdupq_n_s32(0));
+            wide_i = vminq_s32(wide_i, vdupq_n_s32(maxX));
+
+            /* i<<4 | TILEX_LOW_BITS(fx) */
+            wide_lo = vshrq_n_s32(wide_fx, 12);
+            wide_i = vsliq_n_s32(wide_lo, wide_i, 4);
+
+            /* i<<14 */
+            wide_i = vshlq_n_s32(wide_i, 14);
+
+            /* SkClampMax(((f + one)) >> 16, max) */
+            wide_fx1 = vaddq_s32(wide_fx, wide_one);
+            wide_fx1 = vmaxq_s32(vshrq_n_s32(wide_fx1,16), vdupq_n_s32(0));
+            wide_fx1 = vminq_s32(wide_fx1, vdupq_n_s32(maxX));
+
+            /* final combination */
+            wide_i = vorrq_s32(wide_i, wide_fx1);
+
+            vst1q_u32(xy, vreinterpretq_u32_s32(wide_i));
+
+    #if 0
+            /* having a verification hook is a good idea */
+            /* use debug_fx, debug_fx+dx, etc. */
+
+            for (int i=0;i<4;i++) {
+            uint32_t want = PACK_FILTER_X_NAME(debug_fx, maxX, one PREAMBLE_ARG_X);
+                    if (xy[i] != want)
+                {
+                /* print a nastygram */
+                SkDebugf("clamp-filter-scale fails\n");
+                SkDebugf("got %08x want %08x\n", xy[i], want);
+                SkDebugf("fx %08x debug_fx %08x dx %08x done %d\n",
+                fx, debug_fx, dx, count_done);
+                SkDebugf(" maxX %08x one %08x\n", maxX, one);
+
+                }
+            debug_fx += dx;
+            count_done++;
+            }
+    #endif
+            wide_fx += vdupq_n_s32(dx+dx+dx+dx);
+            fx += dx+dx+dx+dx;
+            xy += 4;
+            count -= 4;
+        }
+    }
+   
+    while (--count >= 0) {
+        *xy++ = PACK_FILTER_X_NAME(fx, maxX, one PREAMBLE_ARG_X);
+        fx += dx;
+    }
+}
+
+static void AFFINE_FILTER_NAME(const SkBitmapProcState& s,
+                               uint32_t xy[], int count, int x, int y) {
+    SkASSERT(s.fInvType & SkMatrix::kAffine_Mask);
+    SkASSERT((s.fInvType & ~(SkMatrix::kTranslate_Mask |
+                             SkMatrix::kScale_Mask |
+                             SkMatrix::kAffine_Mask)) == 0);
+    
+    PREAMBLE(s);
+    SkPoint srcPt;
+    s.fInvProc(*s.fInvMatrix,
+               SkIntToScalar(x) + SK_ScalarHalf,
+               SkIntToScalar(y) + SK_ScalarHalf, &srcPt);
+    
+    SkFixed oneX = s.fFilterOneX;
+    SkFixed oneY = s.fFilterOneY;
+    SkFixed fx = SkScalarToFixed(srcPt.fX) - (oneX >> 1);
+    SkFixed fy = SkScalarToFixed(srcPt.fY) - (oneY >> 1);
+    SkFixed dx = s.fInvSx;
+    SkFixed dy = s.fInvKy;
+    unsigned maxX = s.fBitmap->width() - 1;
+    unsigned maxY = s.fBitmap->height() - 1;
+    
+    if (count >= 4) {
+        int32x4_t wide_one, wide_i, wide_lo;
+        int32x4_t wide_dx, wide_fx, wide_onex, wide_fx1;
+        int32x4_t wide_dy, wide_fy, wide_oney, wide_fy1;
+
+        /* need side-by-side registers for vst2.32 tricks */
+        register int32x4_t wide_x asm("q1");
+        register int32x4_t wide_y asm("q0");
+
+    #undef	AFFINE_DEBUG
+    #if	defined(AFFINE_DEBUG)
+        SkFixed fyp = fy;
+        SkFixed fxp = fx;
+        uint32_t *xyp = xy;
+        int count_done = 0;
+    #endif
+
+        wide_fx = vdupq_n_s32(fx);
+        wide_fx = vsetq_lane_s32(fx+dx, wide_fx, 1);
+        wide_fx = vsetq_lane_s32(fx+dx+dx, wide_fx, 2);
+        wide_fx = vsetq_lane_s32(fx+dx+dx+dx, wide_fx, 3);
+        wide_dx = vdupq_n_s32(dx);
+
+        wide_fy = vdupq_n_s32(fy);
+        wide_fy = vsetq_lane_s32(fy+dy, wide_fy, 1);
+        wide_fy = vsetq_lane_s32(fy+dy+dy, wide_fy, 2);
+        wide_fy = vsetq_lane_s32(fy+dy+dy+dy, wide_fy, 3);
+        wide_dy = vdupq_n_s32(dy);
+
+        wide_onex = vdupq_n_s32(oneX);
+        wide_oney = vdupq_n_s32(oneY);
+
+        while (count >= 4) {
+            /* do the X side, then the Y side, then interleave them */
+
+            /* original expands to: 
+             * unsigned i = SkClampMax((f) >> 16, max);
+             * i = (i << 4) | (((f) >> 12) & 0xF);
+             * return (i << 14) | (SkClampMax(((f + one)) >> 16, max));
+             */
+
+            /* i = SkClampMax(f>>16, maxX) */
+            wide_i = vmaxq_s32(vshrq_n_s32(wide_fx,16), vdupq_n_s32(0));
+            wide_i = vminq_s32(wide_i, vdupq_n_s32(maxX));
+
+            /* i<<4 | TILEX_LOW_BITS(fx) */
+            wide_lo = vshrq_n_s32(wide_fx, 12);
+            wide_i = vsliq_n_s32(wide_lo, wide_i, 4);
+
+            /* i<<14 */
+            wide_i = vshlq_n_s32(wide_i, 14);
+
+            /* SkClampMax(((f + one)) >> 16, max) */
+            wide_fx1 = vaddq_s32(wide_fx, wide_onex);
+            wide_fx1 = vmaxq_s32(vshrq_n_s32(wide_fx1,16), vdupq_n_s32(0));
+            wide_fx1 = vminq_s32(wide_fx1, vdupq_n_s32(maxX));
+
+            /* final combination */
+            wide_x = vorrq_s32(wide_i, wide_fx1);
+
+            /* And now the Y side */
+
+            /* i = SkClampMax(f>>16, maxX) */
+            wide_i = vmaxq_s32(vshrq_n_s32(wide_fy,16), vdupq_n_s32(0));
+            wide_i = vminq_s32(wide_i, vdupq_n_s32(maxY));
+
+            /* i<<4 | TILEX_LOW_BITS(fx) */
+            wide_lo = vshrq_n_s32(wide_fy, 12);
+            wide_i = vsliq_n_s32(wide_lo, wide_i, 4);
+
+            /* i<<14 */
+            wide_i = vshlq_n_s32(wide_i, 14);
+
+            /* SkClampMax(((f + one)) >> 16, max) */
+            wide_fy1 = vaddq_s32(wide_fy, wide_oney);
+            wide_fy1 = vmaxq_s32(vshrq_n_s32(wide_fy1,16), vdupq_n_s32(0));
+            wide_fy1 = vminq_s32(wide_fy1, vdupq_n_s32(maxY));
+
+            /* final combination */
+            wide_y = vorrq_s32(wide_i, wide_fy1);
+
+            /* interleave as YXYXYXYX as part of the storing */
+                asm ("vst2.32	{q0-q1},[%2]  /* y=%q0 x=%q1 */"
+                :
+                : "w" (wide_y), "w" (wide_x), "r" (xy)
+                );
+
+    #if	defined(AFFINE_DEBUG)
+            /* make sure we're good here -- check the 4 we just output */
+            for (int i = 0; i<4;i++) {
+            uint32_t val;
+            val = PACK_FILTER_Y_NAME(fyp, maxY, oneY PREAMBLE_ARG_Y);
+            if (val != xy[i*2+0]) {
+                /* print a nastygram */
+                SkDebugf("clamp-filter-affine fails\n");
+                SkDebugf("[bad-y] got %08x want %08x\n", xy[i*2+0], val);
+                SkDebugf("fy %08x fxp %08x fyp %08x dx %08x dy %08x done %d\n",
+                fy, fxp, fyp, dx, dy, count_done);
+                SkDebugf(" maxY %08x oneY %08x\n", maxY, oneY);
+                }
+            val = PACK_FILTER_X_NAME(fxp, maxX, oneX PREAMBLE_ARG_X);
+            if (val != xy[i*2+1]) {
+                /* print a nastygram */
+                SkDebugf("clamp-filter-affine fails\n");
+                SkDebugf("[bad-x] got %08x want %08x\n", xy[i*2+1], val);
+                SkDebugf("fx %08x fxp %08x fyp %08x dx %08x dy %08x done %d\n",
+                fx, fxp, fyp, dx, dy, count_done);
+                SkDebugf(" maxX %08x one %08x\n", maxX, oneX);
+            }
+            fyp += dy;
+            fxp += dx;
+            count_done++;
+            }
+    #endif
+
+            wide_fx += vdupq_n_s32(dx+dx+dx+dx);
+            fx += dx+dx+dx+dx;
+            wide_fy += vdupq_n_s32(dy+dy+dy+dy);
+            fy += dy+dy+dy+dy;
+            xy += 8;		/* 4 x's, 4 y's */
+            count -= 4;
+        }
+    }
+
+    while (--count >= 0) {
+        /* NB: writing Y/X */
+        *xy++ = PACK_FILTER_Y_NAME(fy, maxY, oneY PREAMBLE_ARG_Y);
+        fy += dy;
+        *xy++ = PACK_FILTER_X_NAME(fx, maxX, oneX PREAMBLE_ARG_X);
+        fx += dx;
+    }
+}
+
+static void PERSP_FILTER_NAME(const SkBitmapProcState& s,
+                              uint32_t* SK_RESTRICT xy, int count,
+                              int x, int y) {
+    SkASSERT(s.fInvType & SkMatrix::kPerspective_Mask);
+
+    PREAMBLE(s);
+    unsigned maxX = s.fBitmap->width() - 1;
+    unsigned maxY = s.fBitmap->height() - 1;
+    SkFixed oneX = s.fFilterOneX;
+    SkFixed oneY = s.fFilterOneY;
+
+    SkPerspIter   iter(*s.fInvMatrix,
+                       SkIntToScalar(x) + SK_ScalarHalf,
+                       SkIntToScalar(y) + SK_ScalarHalf, count);
+
+    while ((count = iter.next()) != 0) {
+        const SkFixed* SK_RESTRICT srcXY = iter.getXY();
+
+        if (count >= 4) {
+            int32x4_t wide_one, wide_i, wide_lo;
+            int32x4_t wide_fx1;
+            int32x4_t wide_fy1;
+            int32x4_t wide_x, wide_y;
+
+            /* need side-by-side regs for vld2/vst2 tricks */
+            register int32x4_t wide_first asm ("q0");
+            register int32x4_t wide_second asm ("q1");
+
+            while (count >= 4) {
+                /* RBE: it's good, but:
+                 * -- we spill a constant that could be easily regnerated
+                 *    [perhaps tweak gcc's NEON constant costs?]
+                 */
+
+                /* load src:  x-y-x-y-x-y-x-y */
+                asm ("vld2.32	{q0-q1},[%2]  /* x=%q0 y=%q1 */"
+                : "=w" (wide_first), "=w" (wide_second)
+                : "r" (srcXY));
+
+                wide_x = wide_first;
+                wide_y = wide_second;
+
+                /* do the X side, then the Y side, then interleave them */
+
+                wide_x = vsubq_s32(wide_x, vdupq_n_s32 (oneX>>1));
+
+                /* original expands to: 
+                 * unsigned i = SkClampMax((f) >> 16, max);
+                 * i = (i << 4) | (((f) >> 12) & 0xF);
+                 * return (i << 14) | (SkClampMax(((f + one)) >> 16, max));
+                 */
+
+                /* i = SkClampMax(f>>16, maxX) */
+                wide_i = vmaxq_s32 (vshrq_n_s32 (wide_x, 16), vdupq_n_s32 (0));
+                wide_i = vminq_s32 (wide_i, vdupq_n_s32 (maxX));
+
+                /* i<<4 | TILEX_LOW_BITS(fx) */
+                wide_lo = vshrq_n_s32 (wide_x, 12);
+                wide_i = vsliq_n_s32 (wide_lo, wide_i, 4);
+
+                /* i<<14 */
+                wide_i = vshlq_n_s32 (wide_i, 14);
+
+                /* SkClampMax(((f + one)) >> 16, max) */
+                wide_fx1 = vaddq_s32 (wide_x, vdupq_n_s32(oneX));
+                wide_fx1 = vmaxq_s32 (vshrq_n_s32 (wide_fx1, 16), vdupq_n_s32 (0));
+                wide_fx1 = vminq_s32 (wide_fx1, vdupq_n_s32 (maxX));
+
+                /* final combination */
+                wide_x = vorrq_s32 (wide_i, wide_fx1);
+
+
+                /* And now the Y side */
+
+                wide_y = vsubq_s32(wide_y, vdupq_n_s32 (oneY>>1));
+
+                /* i = SkClampMax(f>>16, maxX) */
+                wide_i = vmaxq_s32 (vshrq_n_s32 (wide_y, 16), vdupq_n_s32 (0));
+                wide_i = vminq_s32 (wide_i, vdupq_n_s32 (maxY));
+
+                /* i<<4 | TILEX_LOW_BITS(fx) */
+                wide_lo = vshrq_n_s32 (wide_y, 12);
+                wide_i = vsliq_n_s32 (wide_lo, wide_i, 4);
+
+                /* i<<14 */
+                wide_i = vshlq_n_s32 (wide_i, 14);
+
+                /* SkClampMax(((f + one)) >> 16, max) */
+                wide_fy1 = vaddq_s32 (wide_y, vdupq_n_s32(oneY));
+                wide_fy1 = vmaxq_s32 (vshrq_n_s32 (wide_fy1, 16), vdupq_n_s32 (0));
+                wide_fy1 = vminq_s32 (wide_fy1, vdupq_n_s32 (maxY));
+
+                /* final combination */
+                wide_y = vorrq_s32 (wide_i, wide_fy1);
+
+                /* switch them around; have to do it this way to get them
+                 * in the proper registers to match our instruction */
+
+                /* wide_x/wide_y are fixed regs, in wrong order; swap 'em */
+                wide_first = wide_y;
+                wide_second = wide_x;
+
+                /* iteration bookkeeping, ahead of the asm() for scheduling */
+                srcXY += 2*4;
+                count -= 4;
+
+                /* store interleaved as y-x-y-x-y-x-y-x (NB != read order) */
+                    asm ("vst2.32	{q0-q1},[%2]  /* y=%q0 x=%q1 */"
+                    :
+                    : "w" (wide_first), "w" (wide_second), "r" (xy));
+
+                /* on to the next iteration */
+                /* count, srcXY are handled above */
+                xy += 2*4;
+            }
+        }
+
+        /* was do-while; NEON code invalidates original count>0 assumption */
+        while (--count >= 0) { 
+	    /* NB: we read x/y, we write y/x */
+            *xy++ = PACK_FILTER_Y_NAME(srcXY[1] - (oneY >> 1), maxY,
+                                       oneY PREAMBLE_ARG_Y);
+            *xy++ = PACK_FILTER_X_NAME(srcXY[0] - (oneX >> 1), maxX,
+                                       oneX PREAMBLE_ARG_X);
+            srcXY += 2;
+        }
+    }
+}
+
+static SkBitmapProcState::MatrixProc MAKENAME(_Procs)[] = {
+    SCALE_NOFILTER_NAME,
+    SCALE_FILTER_NAME,
+    AFFINE_NOFILTER_NAME,
+    AFFINE_FILTER_NAME,
+    PERSP_NOFILTER_NAME,
+    PERSP_FILTER_NAME
+};
+
+#undef MAKENAME
+#undef TILEX_PROCF
+#undef TILEY_PROCF
+#ifdef CHECK_FOR_DECAL
+    #undef CHECK_FOR_DECAL
+#endif
+
+#undef SCALE_NOFILTER_NAME
+#undef SCALE_FILTER_NAME
+#undef AFFINE_NOFILTER_NAME
+#undef AFFINE_FILTER_NAME
+#undef PERSP_NOFILTER_NAME
+#undef PERSP_FILTER_NAME
+
+#undef PREAMBLE
+#undef PREAMBLE_PARAM_X
+#undef PREAMBLE_PARAM_Y
+#undef PREAMBLE_ARG_X
+#undef PREAMBLE_ARG_Y
+
+#undef TILEX_LOW_BITS
+#undef TILEY_LOW_BITS
diff --git a/src/core/SkBitmapProcState_matrix_repeat.h b/src/core/SkBitmapProcState_matrix_repeat.h
new file mode 100644
index 0000000000..2697db6a9f
--- /dev/null
+++ b/src/core/SkBitmapProcState_matrix_repeat.h
@@ -0,0 +1,580 @@
+/* NEON optimized code (C) COPYRIGHT 2009 Motorola */
+/*
+ * Modifications done in-house at Motorola 
+ *
+ * this is a clone of SkBitmapProcState_matrix.h
+ * and has been tuned to work with the NEON unit.
+ *
+ * Still going back and forth between whether this approach
+ * (clone the entire SkBitmapProcState_matrix.h file or
+ * if I should put just the modified routines in here and
+ * then use a construct like #define DONT_DO_THIS_FUNCTION or
+ * something like that...
+ *
+ * This is for the RepeatX_RepeatY part of the world
+ */
+
+
+#if	!defined(__ARM_HAVE_NEON)
+#error	this file can be used only when the NEON unit is enabled
+#endif
+
+#include <arm_neon.h>
+
+/*
+ * This has been modified on the knowledge that (at the time)
+ * we had the following macro definitions in the parent file
+ *
+ * #define MAKENAME(suffix)        RepeatX_RepeatY ## suffix
+ * #define TILEX_PROCF(fx, max)    (((fx) & 0xFFFF) * ((max) + 1) >> 16)
+ * #define TILEY_PROCF(fy, max)    (((fy) & 0xFFFF) * ((max) + 1) >> 16)
+ * #define TILEX_LOW_BITS(fx, max) ((((fx) & 0xFFFF) * ((max) + 1) >> 12) & 0xF)
+ * #define TILEY_LOW_BITS(fy, max) ((((fy) & 0xFFFF) * ((max) + 1) >> 12) & 0xF)
+ */
+
+/* SkClampMax(val,max) -- bound to 0..max */
+
+#define SCALE_NOFILTER_NAME     MAKENAME(_nofilter_scale)
+#define SCALE_FILTER_NAME       MAKENAME(_filter_scale)
+#define AFFINE_NOFILTER_NAME    MAKENAME(_nofilter_affine)
+#define AFFINE_FILTER_NAME      MAKENAME(_filter_affine)
+#define PERSP_NOFILTER_NAME     MAKENAME(_nofilter_persp)
+#define PERSP_FILTER_NAME       MAKENAME(_filter_persp)
+
+#define PACK_FILTER_X_NAME  MAKENAME(_pack_filter_x)
+#define PACK_FILTER_Y_NAME  MAKENAME(_pack_filter_y)
+
+#ifndef PREAMBLE
+    #define PREAMBLE(state)
+    #define PREAMBLE_PARAM_X
+    #define PREAMBLE_PARAM_Y
+    #define PREAMBLE_ARG_X
+    #define PREAMBLE_ARG_Y
+#endif
+
+static void SCALE_NOFILTER_NAME(const SkBitmapProcState& s,
+                                uint32_t xy[], int count, int x, int y) {
+    SkASSERT((s.fInvType & ~(SkMatrix::kTranslate_Mask |
+                             SkMatrix::kScale_Mask)) == 0);
+
+    PREAMBLE(s);
+    // we store y, x, x, x, x, x
+
+    const unsigned maxX = s.fBitmap->width() - 1;
+    SkFixed fx;
+    {
+        SkPoint pt;
+        s.fInvProc(*s.fInvMatrix, SkIntToScalar(x) + SK_ScalarHalf,
+                                  SkIntToScalar(y) + SK_ScalarHalf, &pt);
+        fx = SkScalarToFixed(pt.fY);
+        const unsigned maxY = s.fBitmap->height() - 1;
+        *xy++ = TILEY_PROCF(fx, maxY);
+        fx = SkScalarToFixed(pt.fX);
+    }
+    
+    if (0 == maxX) {
+        // all of the following X values must be 0
+        memset(xy, 0, count * sizeof(uint16_t));
+        return;
+    }
+
+    const SkFixed dx = s.fInvSx;
+
+#ifdef CHECK_FOR_DECAL
+    // test if we don't need to apply the tile proc
+    if ((unsigned)(fx >> 16) <= maxX &&
+        (unsigned)((fx + dx * (count - 1)) >> 16) <= maxX) {
+        decal_nofilter_scale(xy, fx, dx, count);
+    } else
+#endif
+    {
+        int i;
+
+#if	defined(__ARM_HAVE_NEON)
+	/* RBE: very much like done in decal_nofilter ,
+	 * but some processing of the 'fx' information 
+         * TILEX_PROCF(fx, max)    (((fx) & 0xFFFF) * ((max) + 1) >> 16)
+	 */
+	if (count >= 8)
+	{
+	    /* SkFixed is 16.16 fixed point */
+	    SkFixed dx2 = dx+dx;
+	    SkFixed dx4 = dx2+dx2;
+	    SkFixed dx8 = dx4+dx4;
+
+	    /* now build fx/fx+dx/fx+2dx/fx+3dx */
+	    SkFixed fx1, fx2, fx3;
+	    int32x2_t lower, upper;
+	    int32x4_t lbase, hbase;
+	    int16_t *dst16 = (int16_t *)xy;
+
+	    fx1 = fx+dx;
+	    fx2 = fx1+dx;
+	    fx3 = fx2+dx;
+
+#if 1
+	    /* avoid the 'lbase unitialized' warning */
+	    lbase = vdupq_n_s32(fx);
+#else
+	    lbase = vsetq_lane_s32(fx, lbase, 0);
+#endif
+	    lbase = vsetq_lane_s32(fx1, lbase, 1);
+	    lbase = vsetq_lane_s32(fx2, lbase, 2);
+	    lbase = vsetq_lane_s32(fx3, lbase, 3);
+	    hbase = vaddq_s32(lbase, vdupq_n_s32(dx4));
+
+	    /* store & bump */
+	    do
+	    {
+	        int32x4_t lout;
+		int32x4_t hout;
+		int16x8_t hi16;
+
+         	/* TILEX_PROCF(fx, max) (((fx)&0xFFFF)*((max)+1)>> 16) */
+		/* mask to low 16 [would like to use uzp tricks) */
+	        lout = vandq_s32(lbase, vdupq_n_s32(0xffff));
+	        hout = vandq_s32(hbase, vdupq_n_s32(0xffff));
+		/* bare multiplication, not SkFixedMul */
+		lout = vmulq_s32(lout, vdupq_n_s32(maxX+1));
+		hout = vmulq_s32(hout, vdupq_n_s32(maxX+1));
+
+		/* extraction, using uzp */
+		/* this is ok -- we want all hi(lout)s then all hi(hout)s */
+		asm ("vuzpq.16 %q0, %q1" : "+w" (lout), "+w" (hout));
+		hi16 = vreinterpretq_s16_s32(hout);
+		vst1q_s16(dst16, hi16);
+
+		/* bump our base on to the next */
+		lbase = vaddq_s32 (lbase, vdupq_n_s32(dx8));
+		hbase = vaddq_s32 (hbase, vdupq_n_s32(dx8));
+		dst16 += 8;
+		count -= 8;
+		fx += dx8;
+	    } while (count >= 8);
+	    xy = (uint32_t *) dst16;
+	}
+#else
+	/* simple, portable way of looking at 4 at a crack;
+	 * so gets some loop unrolling, but not full SIMD speed
+	 */
+        for (i = (count >> 2); i > 0; --i) {
+            unsigned a, b;
+            a = TILEX_PROCF(fx, maxX); fx += dx;
+            b = TILEX_PROCF(fx, maxX); fx += dx;
+#ifdef SK_CPU_BENDIAN
+            *xy++ = (a << 16) | b;
+#else
+            *xy++ = (b << 16) | a;
+#endif
+            a = TILEX_PROCF(fx, maxX); fx += dx;
+            b = TILEX_PROCF(fx, maxX); fx += dx;
+#ifdef SK_CPU_BENDIAN
+            *xy++ = (a << 16) | b;
+#else
+            *xy++ = (b << 16) | a;
+#endif
+        }
+	/* loop doesn't adjust count */
+	count -= (count>>2);
+#endif
+        uint16_t* xx = (uint16_t*)xy;
+        for (i = count; i > 0; --i) {
+            *xx++ = TILEX_PROCF(fx, maxX); fx += dx;
+        }
+    }
+}
+
+// note: we could special-case on a matrix which is skewed in X but not Y.
+// this would require a more general setup thatn SCALE does, but could use
+// SCALE's inner loop that only looks at dx
+
+
+static void AFFINE_NOFILTER_NAME(const SkBitmapProcState& s,
+                                 uint32_t xy[], int count, int x, int y) {
+    SkASSERT(s.fInvType & SkMatrix::kAffine_Mask);
+    SkASSERT((s.fInvType & ~(SkMatrix::kTranslate_Mask |
+                             SkMatrix::kScale_Mask |
+                             SkMatrix::kAffine_Mask)) == 0);
+    
+    PREAMBLE(s);
+    SkPoint srcPt;
+    s.fInvProc(*s.fInvMatrix,
+               SkIntToScalar(x) + SK_ScalarHalf,
+               SkIntToScalar(y) + SK_ScalarHalf, &srcPt);
+    
+    SkFixed fx = SkScalarToFixed(srcPt.fX);
+    SkFixed fy = SkScalarToFixed(srcPt.fY);
+    SkFixed dx = s.fInvSx;
+    SkFixed dy = s.fInvKy;
+    int maxX = s.fBitmap->width() - 1;
+    int maxY = s.fBitmap->height() - 1;
+
+#if 1
+    int ocount = count;
+    uint32_t *oxy = xy;
+    SkFixed bfx = fx, bfy=fy, bdx=dx, bdy=dy;
+#endif
+
+#if	defined(__ARM_HAVE_NEON)
+
+	if (0) { extern void rbe(void); rbe(); }
+
+	/* RBE: benchmarks show this eats up time; can we neonize it? */
+	/* RBE: very much like done in decal_nofilter ,
+	 * but some processing of the 'fx' information 
+         * TILEX_PROCF(fx, max)    (((fx) & 0xFFFF) * ((max) + 1) >> 16)
+	 */
+	if (count >= 4)
+	{
+	    /* SkFixed is 16.16 fixed point */
+	    SkFixed dx4 = dx*4;
+	    SkFixed dy4 = dy*4;
+
+	    /* now build fx/fx+dx/fx+2dx/fx+3dx */
+	    int32x2_t lower, upper;
+	    int32x4_t xbase, ybase;
+	    int16_t *dst16 = (int16_t *)xy;
+
+	    /* synthesize 4x for both X and Y */
+#if 1
+	    /* avoid the xbase unitialized warning */
+	    xbase = vdupq_n_s32(fx);
+#else
+	    xbase = vsetq_lane_s32(fx, xbase, 0);
+#endif
+	    xbase = vsetq_lane_s32(fx+dx, xbase, 1);
+	    xbase = vsetq_lane_s32(fx+dx+dx, xbase, 2);
+	    xbase = vsetq_lane_s32(fx+dx+dx+dx, xbase, 3);
+
+#if 1
+	    /* avoid the xbase unitialized warning */
+	    ybase = vdupq_n_s32(fy);
+#else
+	    ybase = vsetq_lane_s32(fy, ybase, 0);
+#endif
+	    ybase = vsetq_lane_s32(fy+dy, ybase, 1);
+	    ybase = vsetq_lane_s32(fy+dy+dy, ybase, 2);
+	    ybase = vsetq_lane_s32(fy+dy+dy+dy, ybase, 3);
+
+	    /* store & bump */
+	    do
+	    {
+	        int32x4_t xout;
+		int32x4_t yout;
+		int16x8_t hi16;
+
+         	/* TILEX_PROCF(fx, max) (((fx)&0xFFFF)*((max)+1)>> 16) */
+		/* mask to low 16 [would like to use uzp tricks) */
+	        xout = vandq_s32(xbase, vdupq_n_s32(0xffff));
+	        yout = vandq_s32(ybase, vdupq_n_s32(0xffff));
+		/* bare multiplication, not SkFixedMul */
+		xout = vmulq_s32(xout, vdupq_n_s32(maxX+1));
+		yout = vmulq_s32(yout, vdupq_n_s32(maxY+1));
+
+		/* put hi16 from xout over low16 from yout */
+		yout = vsriq_n_s32(yout, xout, 16);
+
+		/* and then yout has the interleaved upper 16's */
+		hi16 = vreinterpretq_s16_s32(yout);
+		vst1q_s16(dst16, hi16);
+
+		/* bump preserved base & on to the next */
+		xbase = vaddq_s32 (xbase, vdupq_n_s32(dx4));
+		ybase = vaddq_s32 (ybase, vdupq_n_s32(dy4));
+		dst16 += 8;	/* 8 x16 aka 4x32 */
+		count -= 4;
+		fx += dx4;
+		fy += dy4;
+	    } while (count >= 4);
+	    xy = (uint32_t *) dst16;
+	}
+#if 0
+    /* diagnostics... see whether we agree with the NEON code */
+    int bad = 0;
+    uint32_t *myxy = oxy;
+    int myi = (-1);
+    SkFixed ofx = bfx, ofy= bfy, odx= bdx, ody= bdy;
+    for (myi = ocount; myi > 0; --myi) {
+	uint32_t val = (TILEY_PROCF(ofy, maxY) << 16) | TILEX_PROCF(ofx, maxX);
+	if (val != *myxy++) {
+		bad++;
+		break;
+	}
+        ofx += odx; ofy += ody;
+    }
+    if (bad) {
+	SkDebugf("repeat-nofilter-affine fails\n");
+	SkDebugf("count %d myi %d\n", ocount, myi);
+	SkDebugf(" bfx %08x, bdx %08x, bfy %08x bdy %08x\n",
+			bfx, bdx, bfy, bdy);
+	SkDebugf("maxX %08x maxY %08x\n", maxX, maxY);
+    }
+#endif
+#endif
+
+    for (int i = count; i > 0; --i) {
+	/* fx, fy, dx, dy are all 32 bit 16.16 fixed point */
+	/* (((fx) & 0xFFFF) * ((max) + 1) >> 16) */
+        *xy++ = (TILEY_PROCF(fy, maxY) << 16) | TILEX_PROCF(fx, maxX);
+        fx += dx; fy += dy;
+    }
+}
+
+static void PERSP_NOFILTER_NAME(const SkBitmapProcState& s,
+                                uint32_t* SK_RESTRICT xy,
+                                int count, int x, int y) {
+    SkASSERT(s.fInvType & SkMatrix::kPerspective_Mask);
+    
+    PREAMBLE(s);
+    int maxX = s.fBitmap->width() - 1;
+    int maxY = s.fBitmap->height() - 1;
+    
+    SkPerspIter   iter(*s.fInvMatrix,
+                       SkIntToScalar(x) + SK_ScalarHalf,
+                       SkIntToScalar(y) + SK_ScalarHalf, count);
+    
+    while ((count = iter.next()) != 0) {
+        const SkFixed* SK_RESTRICT srcXY = iter.getXY();
+
+#if	defined(__ARM_HAVE_NEON)
+	/* RBE: */
+	/* TILEX_PROCF(fx, max) (((fx) & 0xFFFF) * ((max) + 1) >> 16) */
+	/* it's a little more complicated than what I did for the
+	 * clamp case -- where I could immediately snip to the top
+	 * 16 bits and do my min/max games there.
+	 * ... might only be able to get 4x unrolling here
+	 */
+
+	/* vld2 to get a set of 32x4's ... */
+	/* do the tile[xy]_procf operations */
+	/* which includes doing vuzp to get hi16's */
+	/* store it */
+	/* -- inner loop (other than vld2) can be had from above */
+
+	/* srcXY is a batch of 32 bit numbers X0,Y0,X1,Y1...
+	 * but we immediately discard the low 16 bits...
+	 * so what we're going to do is vld4, which will give us
+	 * xlo,xhi,ylo,yhi distribution and we can ignore the 'lo'
+	 * parts....
+	 */
+	if (0) { extern void rbe(void); rbe(); }
+	if (count >= 8) {
+	    int32_t *mysrc = (int32_t *) srcXY;
+	    int16_t *mydst = (int16_t *) xy;
+	    do {
+	    	register int32x4_t x asm("q0");
+	    	register int32x4_t y asm("q1");
+	    	register int32x4_t x2 asm("q2");
+	    	register int32x4_t y2 asm("q3");
+
+		int16x8_t hi;
+		int16x8_t hi2;
+
+		/* read array of x,y,x,y,x,y */
+	        /* vld2 does the de-interleaving for us */
+	        /* dependent on register assignments above */
+    	        asm ("vld2.32	{q0-q1},[%2]  /* x=%q0 y=%q1 */"
+		    : "=w" (x), "=w" (y)
+		    : "r" (mysrc)
+		    );
+
+		/* offset == 256 bits == 32 bytes == 8 longs */
+    	        asm ("vld2.32	{q2-q3},[%2,#32]  /* x=%q0 y=%q1 */"
+		    : "=w" (x2), "=w" (y2)
+		    : "r" (mysrc)
+		    );
+
+         	/* TILEX_PROCF(fx, max) (((fx)&0xFFFF)*((max)+1)>> 16) */
+		/* mask to low 16 [would like to use uzp tricks) */
+		/* bare multiplication, not SkFixedMul */
+	        x = vandq_s32(x, vdupq_n_s32(0xffff));
+		x = vmulq_s32(x, vdupq_n_s32(maxX+1));
+	        y = vandq_s32(y, vdupq_n_s32(0xffff));
+		y = vmulq_s32(y, vdupq_n_s32(maxY+1));
+
+	        x2 = vandq_s32(x2, vdupq_n_s32(0xffff));
+		x2 = vmulq_s32(x2, vdupq_n_s32(maxX+1));
+	        y2 = vandq_s32(y2, vdupq_n_s32(0xffff));
+		y2 = vmulq_s32(y2, vdupq_n_s32(maxY+1));
+
+		/* now collect interleaved high 16's */
+		/* (hi-x, hi-y)4  (hi-x2; hi-y2)4 */
+
+		/* extraction, using uzp, leaves hi16's in y */
+		y = vsriq_n_s32(y, x, 16);
+		hi = vreinterpretq_s16_s32(y);
+		vst1q_s16(mydst, hi);
+
+		/* and get second 8 bytes out */
+		y2 = vsriq_n_s32(y2, x2, 16);
+		hi2 = vreinterpretq_s16_s32(y2);
+		vst1q_s16(mydst+8, hi2);
+
+		/* XXX: gcc isn't interleaving these with the NEON ops
+		 * but i think that all the scoreboarding works out */
+		count -= 8;	/* 8 iterations */
+		mysrc += 16;	/* 16 longs */
+		mydst += 16;	/* 16 shorts, aka 8 longs */
+	    } while (count >= 8);
+	    /* get xy and srcXY fixed up */
+	    srcXY = (const SkFixed *) mysrc;
+	    xy = (uint32_t *) mydst;
+	}
+#endif
+        while (--count >= 0) {
+            *xy++ = (TILEY_PROCF(srcXY[1], maxY) << 16) |
+                     TILEX_PROCF(srcXY[0], maxX);
+            srcXY += 2;
+        }
+    }
+}
+
+//////////////////////////////////////////////////////////////////////////////
+
+static inline uint32_t PACK_FILTER_Y_NAME(SkFixed f, unsigned max,
+                                          SkFixed one PREAMBLE_PARAM_Y) {
+    unsigned i = TILEY_PROCF(f, max);
+    i = (i << 4) | TILEY_LOW_BITS(f, max);
+    return (i << 14) | (TILEY_PROCF((f + one), max));
+}
+
+static inline uint32_t PACK_FILTER_X_NAME(SkFixed f, unsigned max,
+                                          SkFixed one PREAMBLE_PARAM_X) {
+    unsigned i = TILEX_PROCF(f, max);
+    i = (i << 4) | TILEX_LOW_BITS(f, max);
+    return (i << 14) | (TILEX_PROCF((f + one), max));
+}
+
+static void SCALE_FILTER_NAME(const SkBitmapProcState& s,
+                              uint32_t xy[], int count, int x, int y) {
+    SkASSERT((s.fInvType & ~(SkMatrix::kTranslate_Mask |
+                             SkMatrix::kScale_Mask)) == 0);
+    SkASSERT(s.fInvKy == 0);
+
+    PREAMBLE(s);
+    
+    const unsigned maxX = s.fBitmap->width() - 1;
+    const SkFixed one = s.fFilterOneX;
+    const SkFixed dx = s.fInvSx;
+    SkFixed fx;
+
+    {
+        SkPoint pt;
+        s.fInvProc(*s.fInvMatrix, SkIntToScalar(x) + SK_ScalarHalf,
+                                  SkIntToScalar(y) + SK_ScalarHalf, &pt);
+        const SkFixed fy = SkScalarToFixed(pt.fY) - (s.fFilterOneY >> 1);
+        const unsigned maxY = s.fBitmap->height() - 1;
+        // compute our two Y values up front
+        *xy++ = PACK_FILTER_Y_NAME(fy, maxY, s.fFilterOneY PREAMBLE_ARG_Y);
+        // now initialize fx
+        fx = SkScalarToFixed(pt.fX) - (one >> 1);
+    }
+
+#ifdef CHECK_FOR_DECAL
+    // test if we don't need to apply the tile proc
+    if (dx > 0 &&
+            (unsigned)(fx >> 16) <= maxX &&
+            (unsigned)((fx + dx * (count - 1)) >> 16) < maxX) {
+        decal_filter_scale(xy, fx, dx, count);
+    } else
+#endif
+    {
+        do {
+            *xy++ = PACK_FILTER_X_NAME(fx, maxX, one PREAMBLE_ARG_X);
+            fx += dx;
+        } while (--count != 0);
+    }
+}
+
+static void AFFINE_FILTER_NAME(const SkBitmapProcState& s,
+                               uint32_t xy[], int count, int x, int y) {
+    SkASSERT(s.fInvType & SkMatrix::kAffine_Mask);
+    SkASSERT((s.fInvType & ~(SkMatrix::kTranslate_Mask |
+                             SkMatrix::kScale_Mask |
+                             SkMatrix::kAffine_Mask)) == 0);
+    
+    PREAMBLE(s);
+    SkPoint srcPt;
+    s.fInvProc(*s.fInvMatrix,
+               SkIntToScalar(x) + SK_ScalarHalf,
+               SkIntToScalar(y) + SK_ScalarHalf, &srcPt);
+    
+    SkFixed oneX = s.fFilterOneX;
+    SkFixed oneY = s.fFilterOneY;
+    SkFixed fx = SkScalarToFixed(srcPt.fX) - (oneX >> 1);
+    SkFixed fy = SkScalarToFixed(srcPt.fY) - (oneY >> 1);
+    SkFixed dx = s.fInvSx;
+    SkFixed dy = s.fInvKy;
+    unsigned maxX = s.fBitmap->width() - 1;
+    unsigned maxY = s.fBitmap->height() - 1;
+    
+    do {
+        *xy++ = PACK_FILTER_Y_NAME(fy, maxY, oneY PREAMBLE_ARG_Y);
+        fy += dy;
+        *xy++ = PACK_FILTER_X_NAME(fx, maxX, oneX PREAMBLE_ARG_X);
+        fx += dx;
+    } while (--count != 0);
+}
+
+static void PERSP_FILTER_NAME(const SkBitmapProcState& s,
+                              uint32_t* SK_RESTRICT xy, int count,
+                              int x, int y) {
+    SkASSERT(s.fInvType & SkMatrix::kPerspective_Mask);
+
+    extern void rbe(void);
+    
+    PREAMBLE(s);
+    unsigned maxX = s.fBitmap->width() - 1;
+    unsigned maxY = s.fBitmap->height() - 1;
+    SkFixed oneX = s.fFilterOneX;
+    SkFixed oneY = s.fFilterOneY;
+
+
+    
+    SkPerspIter   iter(*s.fInvMatrix,
+                       SkIntToScalar(x) + SK_ScalarHalf,
+                       SkIntToScalar(y) + SK_ScalarHalf, count);
+
+    while ((count = iter.next()) != 0) {
+        const SkFixed* SK_RESTRICT srcXY = iter.getXY();
+        do {
+            *xy++ = PACK_FILTER_Y_NAME(srcXY[1] - (oneY >> 1), maxY,
+                                       oneY PREAMBLE_ARG_Y);
+            *xy++ = PACK_FILTER_X_NAME(srcXY[0] - (oneX >> 1), maxX,
+                                       oneX PREAMBLE_ARG_X);
+            srcXY += 2;
+        } while (--count != 0);
+    }
+}
+
+static SkBitmapProcState::MatrixProc MAKENAME(_Procs)[] = {
+    SCALE_NOFILTER_NAME,
+    SCALE_FILTER_NAME,
+    AFFINE_NOFILTER_NAME,
+    AFFINE_FILTER_NAME,
+    PERSP_NOFILTER_NAME,
+    PERSP_FILTER_NAME
+};
+
+#undef MAKENAME
+#undef TILEX_PROCF
+#undef TILEY_PROCF
+#ifdef CHECK_FOR_DECAL
+    #undef CHECK_FOR_DECAL
+#endif
+
+#undef SCALE_NOFILTER_NAME
+#undef SCALE_FILTER_NAME
+#undef AFFINE_NOFILTER_NAME
+#undef AFFINE_FILTER_NAME
+#undef PERSP_NOFILTER_NAME
+#undef PERSP_FILTER_NAME
+
+#undef PREAMBLE
+#undef PREAMBLE_PARAM_X
+#undef PREAMBLE_PARAM_Y
+#undef PREAMBLE_ARG_X
+#undef PREAMBLE_ARG_Y
+
+#undef TILEX_LOW_BITS
+#undef TILEY_LOW_BITS
diff --git a/xcode/sampleapp/SampleApp.xcodeproj/project.pbxproj b/xcode/sampleapp/SampleApp.xcodeproj/project.pbxproj
index 50aee209ca..2f89460357 100644
--- a/xcode/sampleapp/SampleApp.xcodeproj/project.pbxproj
+++ b/xcode/sampleapp/SampleApp.xcodeproj/project.pbxproj
@@ -23,6 +23,7 @@
 		00003C9E0EFC233F000FF73A /* SkDOM.cpp in Sources */ = {isa = PBXBuildFile; fileRef = 00003C9A0EFC233F000FF73A /* SkDOM.cpp */; };
 		00003CA10EFC233F000FF73A /* SkXMLParser.cpp in Sources */ = {isa = PBXBuildFile; fileRef = 00003C9D0EFC233F000FF73A /* SkXMLParser.cpp */; };
 		00003CA40EFC235F000FF73A /* SkXMLParser_empty.cpp in Sources */ = {isa = PBXBuildFile; fileRef = 00003CA30EFC235F000FF73A /* SkXMLParser_empty.cpp */; };
+		0009E2201057E96800B0DE6F /* SampleStrokePath.cpp in Sources */ = {isa = PBXBuildFile; fileRef = 0009E21F1057E96800B0DE6F /* SampleStrokePath.cpp */; };
 		000A99820FD97526007E45BD /* SampleArc.cpp in Sources */ = {isa = PBXBuildFile; fileRef = 00A41E4A0EFC312F00C9CBEB /* SampleArc.cpp */; };
 		001B871E1042184D00C84ED4 /* Forth.cpp in Sources */ = {isa = PBXBuildFile; fileRef = 001B871D1042184D00C84ED4 /* Forth.cpp */; };
 		0028847B0EFAB46A0083E387 /* libcore.a in Frameworks */ = {isa = PBXBuildFile; fileRef = 002884510EFAA35C0083E387 /* libcore.a */; };
@@ -75,6 +76,7 @@
 		00AF9B18103CD5EB00CBBCB3 /* SampleDitherBitmap.cpp in Sources */ = {isa = PBXBuildFile; fileRef = 00AF9B17103CD5EB00CBBCB3 /* SampleDitherBitmap.cpp */; };
 		00BB289B104781D00057BF7E /* SampleForth.cpp in Sources */ = {isa = PBXBuildFile; fileRef = 00BB289A104781D00057BF7E /* SampleForth.cpp */; };
 		00C1B809103857A400FA5948 /* SampleFillType.cpp in Sources */ = {isa = PBXBuildFile; fileRef = 0041CE270F00A12400695E8C /* SampleFillType.cpp */; };
+		00EB4593104DBB18002B413E /* ForthTests.cpp in Sources */ = {isa = PBXBuildFile; fileRef = 00EB4592104DBB18002B413E /* ForthTests.cpp */; };
 		00ED55F3104A10EB00F51FF8 /* StdWords.cpp in Sources */ = {isa = PBXBuildFile; fileRef = 00ED55F2104A10EB00F51FF8 /* StdWords.cpp */; };
 		00F53F480FFCFC4D003FA70A /* SampleGradients.cpp in Sources */ = {isa = PBXBuildFile; fileRef = 00C55DA00F8552DC000CAC09 /* SampleGradients.cpp */; };
 		00FF39140FC6ED2C00915187 /* SampleEffects.cpp in Sources */ = {isa = PBXBuildFile; fileRef = 00FF39130FC6ED2C00915187 /* SampleEffects.cpp */; };
@@ -153,6 +155,7 @@
 		00003C9A0EFC233F000FF73A /* SkDOM.cpp */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.cpp.cpp; name = SkDOM.cpp; path = ../../src/xml/SkDOM.cpp; sourceTree = SOURCE_ROOT; };
 		00003C9D0EFC233F000FF73A /* SkXMLParser.cpp */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.cpp.cpp; name = SkXMLParser.cpp; path = ../../src/xml/SkXMLParser.cpp; sourceTree = SOURCE_ROOT; };
 		00003CA30EFC235F000FF73A /* SkXMLParser_empty.cpp */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.cpp.cpp; name = SkXMLParser_empty.cpp; path = ../../src/ports/SkXMLParser_empty.cpp; sourceTree = SOURCE_ROOT; };
+		0009E21F1057E96800B0DE6F /* SampleStrokePath.cpp */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.cpp.cpp; name = SampleStrokePath.cpp; path = ../../samplecode/SampleStrokePath.cpp; sourceTree = SOURCE_ROOT; };
 		001B871D1042184D00C84ED4 /* Forth.cpp */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.cpp.cpp; name = Forth.cpp; path = ../../forth/Forth.cpp; sourceTree = SOURCE_ROOT; };
 		002884490EFAA35C0083E387 /* core.xcodeproj */ = {isa = PBXFileReference; lastKnownFileType = "wrapper.pb-project"; name = core.xcodeproj; path = ../core/core.xcodeproj; sourceTree = SOURCE_ROOT; };
 		002884B40EFAB69F0083E387 /* maccore.xcodeproj */ = {isa = PBXFileReference; lastKnownFileType = "wrapper.pb-project"; name = maccore.xcodeproj; path = ../maccore/maccore.xcodeproj; sourceTree = SOURCE_ROOT; };
@@ -208,6 +211,7 @@
 		00BB289A104781D00057BF7E /* SampleForth.cpp */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.cpp.cpp; name = SampleForth.cpp; path = ../../forth/SampleForth.cpp; sourceTree = SOURCE_ROOT; };
 		00C55DA00F8552DC000CAC09 /* SampleGradients.cpp */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.cpp.cpp; name = SampleGradients.cpp; path = ../../samplecode/SampleGradients.cpp; sourceTree = SOURCE_ROOT; };
 		00D6B5CB0F72DC4300C466B9 /* SampleFuzz.cpp */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.cpp.cpp; name = SampleFuzz.cpp; path = ../../samplecode/SampleFuzz.cpp; sourceTree = SOURCE_ROOT; };
+		00EB4592104DBB18002B413E /* ForthTests.cpp */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.cpp.cpp; name = ForthTests.cpp; path = ../../forth/ForthTests.cpp; sourceTree = SOURCE_ROOT; };
 		00ED55F2104A10EB00F51FF8 /* StdWords.cpp */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.cpp.cpp; name = StdWords.cpp; path = ../../forth/StdWords.cpp; sourceTree = SOURCE_ROOT; };
 		00FF39130FC6ED2C00915187 /* SampleEffects.cpp */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.cpp.cpp; name = SampleEffects.cpp; path = ../../samplecode/SampleEffects.cpp; sourceTree = SOURCE_ROOT; };
 		0156F80307C56A3000C6122B /* Foundation.framework */ = {isa = PBXFileReference; lastKnownFileType = wrapper.framework; name = Foundation.framework; path = /System/Library/Frameworks/Foundation.framework; sourceTree = "<absolute>"; };
@@ -289,6 +293,7 @@
 				0041CE270F00A12400695E8C /* SampleFillType.cpp */,
 				00AF9B17103CD5EB00CBBCB3 /* SampleDitherBitmap.cpp */,
 				00840B74104C69F5005B3EDC /* SampleExtractAlpha.cpp */,
+				0009E21F1057E96800B0DE6F /* SampleStrokePath.cpp */,
 				0041CE280F00A12400695E8C /* SampleFilter.cpp */,
 				0041CE290F00A12400695E8C /* SampleFilter2.cpp */,
 				0041CE2A0F00A12400695E8C /* SampleFontCache.cpp */,
@@ -366,6 +371,7 @@
 			children = (
 				001B871D1042184D00C84ED4 /* Forth.cpp */,
 				00ED55F2104A10EB00F51FF8 /* StdWords.cpp */,
+				00EB4592104DBB18002B413E /* ForthTests.cpp */,
 				00BB289A104781D00057BF7E /* SampleForth.cpp */,
 				2762F66A0FCCCAA2002BD8B4 /* images */,
 				00003C6A0EFC22AD000FF73A /* views */,
@@ -595,6 +601,8 @@
 				00BB289B104781D00057BF7E /* SampleForth.cpp in Sources */,
 				00ED55F3104A10EB00F51FF8 /* StdWords.cpp in Sources */,
 				00840B75104C69F5005B3EDC /* SampleExtractAlpha.cpp in Sources */,
+				00EB4593104DBB18002B413E /* ForthTests.cpp in Sources */,
+				0009E2201057E96800B0DE6F /* SampleStrokePath.cpp in Sources */,
 			);
 			runOnlyForDeploymentPostprocessing = 0;
 		};